Krisp started out as an AI-powered noise reduction service, but now the startup goes beyond that by offering instant transcription of all calls and meetings on your device — and of course, it now also integrates ChatGPT for quick summaries. The open beta starts today.
My experience with transcription usually involves downloading a video or audio file associated with a meeting, provided someone thought to record it (and I know who that person was), then uploading that file to a service like Otter and waiting 10– 15 minutes to process. Compared to the situation a few years ago, it is of course very convenient, but the number of meetings held online has increased. Where I can shave a few minutes or get away from work, I’m usually interested.
Over the years, Krisp has also shown determination by focusing on effective noise reduction in many devices and services. The advance that made this possible was essentially training the system to recognize and isolate human voices, which, while diverse, are also distinct and very different from barking dogs, traffic noise, and so on. For me, Krisp overall just worked.
So when they told me they were doing transcription, I was elated to hear that: it’s a natural extension of a voice-centric platform, and of course they know they can get a pristine voice signal.
Krisp already works more or less behind the scenes, acting as an intermediary for audio sources. So there’s no need “AI Meeting Assistant” integrate with Google, Teams, Zoom or whatever, or be a phantom participant – it only deals with your audio signal before it goes out or before it reaches your ears. The transcription agent doesn’t need to communicate with any services or APIs, it simply identifies incoming and outgoing voices and transcribes them on your device.
On-device transcription is a key differentiator, on the one hand because it is simpler than handling audio files, and on the other hand, because the meeting audio never leaves the computer. I’m not sure for which threat models this is relevant, but I suspect they exist and will become even more prominent in the coming years.
Krisp Nearly Triples Fundraising with $9M Enhance After 2020 Blockbuster Success
The transcript itself is sent directly to the Krisp cloud service, which sort of negates the security benefits of on-device transcription. But if you want to trust another translator, why not this one? Either way, there will also soon be an option to store your transcripts entirely on your device. Krisp CEO and co-founder Davit Baghdasaryan has extensive experience in security and said he is looking forward to implementing the solution. (I suspect it’s simply easier to build a web app than to add significant recent features to multiple mobile and desktop apps).
How about the transcriptions? In my experience, about as good as other AI-based ones, which is better than my notes, but not quite ready for publication. Dropping the audio immediately negates one useful feature offered by other transcriptionists, namely quick checking and editing: click on a suspicious word and it will jump to that point in the audio. At least for now, for interviews where you need to quote carefully and extensively, this may not be the best choice.
But Krisp sees it more as a productivity tool, building a platform for quickly extracting key points from the countless meetings we sometimes schedule. To this end, the company allows transcripts to be quickly submitted for analysis via ChatGPT, which extracts the TL; DR, assigned actions, or anything else that needs to be taken away from the meeting.
Honestly, I never liked these features and never used them, so I can’t comment on the quality here. They are probably as good as any other service that does this, which is to say, useful to people who find such things useful. I try to forget the content of the meeting as quickly as possible, no doubt to the chagrin of my colleagues, but in my defense I simply remembered it wrong. Organized people who have processes for this type of work may or may not find Krisp’s style good for them.
I’m also hesitant to evaluate a service that is essentially an API call to another service that itself is constantly changing. As ChatGPT improves, so will applications based on it – as well as on a hundred other platforms that have suddenly developed the ability to understand natural language.
Krisp is free if you want to try it out – the company makes money through premium enterprise subscriptions, and of course, with the calculations running on your device, scaling is quite straightforward.
It’s worth spending a moment on your Mac or PC to make sure it’s configured correctly, which means its upload is working as expected. I had a few weird issues at first, but I solved them by starting from scratch and only assigning meeting apps and microphones to Krisp so it wouldn’t mix up my music. It’s just a few knobs and switches, so spend five minutes getting it right and soon your transcripts will be coming out so swift it’ll be a little annoying.